Started cubing in Feb, I use cfop w 2 looks OLL e PLL but I started studying for 1 look PLL this week and working on a faster F2L.
These are my stats. Probably I’m not gonna be a professional cuber 🤣 but still proud.
How do you estimate the credit cost of a project once it’s finished, in order to bill the client — or do you use a different method to protect yourselves from excessive costs from Seedance 2, upscaling, 4K, and constant feedback requiring regenerations?
I usually charge a daily fee for motion design work, but maybe for generative AI projects it’s not a good solution, since in one day of video work I can spend up to €200.
Ho 22 anni, e lavoro a Milano.
Attualmente spendo 900€ di affitto: tutti soldi bruciati.
Sto iniziando a pensare che forse ha più senso comprare una casa e spendere anche qualcosina in più (1200-1400), ma di rata del mutuo.
Essendo Under 36 e non avendo nulla intestato credo di poter avere qualche agevolazione tipo il mutuo al 100% (?).
Attualmente fatturo 40-60k annui da forfettario ma sono solo al secondo anno di lavoro a tempo pieno e conto di avere una bella crescita sopratutto nei primi anni.
Ovviamente essendo in forfettario ho un buon netto.
(E ho dei garanti)
Mi sembra un idea troppo bella per essere vera, sicuramente una volta dentro mi renderei conto di aver sopravvalutato tutto. Quindi vorrei chiedervi, cosa non sto considerando? È davvero così conveniente comprare casa per viverci? Magari non ci vivrò per sempre ma almeno parte dei soldi che sto versando adesso per poter vivere a Milano dove ci sono tutte le opportunità finiscono sul mio patrimonio futuro e non sulle tasche di un altro.
Riportatemi coi piedi per terra
For users on plans that include Unlimited generations, it would be great to understand why MCP access is treated differently and whether support for Unlimited Mode through MCP is planned in the future.
Is anyone else experiencing issues with the Magnific MCP server constantly disconnecting from Claude?
The connection works fine initially, but after a few hours (or after a certain number of tasks), Claude suddenly reports that the Magnific MCP server is unavailable and asks me to reconnect it. Once I reconnect, everything works again, but the same thing happens repeatedly.
Connect the Magnific MCP to a client like Claude and the model can read your Spaces’ node graphs, restructure them headless, run workflows, and create new Spaces. It works on your own Spaces and on ones shared with you as editor.
Why it matters: you stop treating prompting as vibes. The model reasons over the actual graph and enforces discipline a chat only setup never does. Consistency stops being a “please remember the character” plea and becomes a wiring problem: you lock the asset as an input edge, and the graph becomes the memory.
The visual iteration stays in the UI where it belongs. The model handles the architecture and the prompt engineering on top. Surprised this isn’t talked about more.
Paste this context once so the model behaves like a Spaces architect instead of a generic prompt writer:
You operate on my Magnific Spaces via the MCP server. You are the graph
architect and prompt engineer, not the visual iterator. I validate images
in the visual.
CAPABILITIES: read graphs (spaces_list / spaces_state / spaces_get_nodes),
restructure headless (spaces_edit), execute (spaces_run), create (spaces_create).
Reason over the REAL graph, not abstractions.
PROMPTING DISCIPLINE (hard rule):
- Every generation call is STATELESS.
- A relative instruction ("more X than before", "same as previous") is emitted
ONLY if the referent image is an explicit input node. Otherwise describe the
target in ABSOLUTE terms. Never reference a generation not present in the call.
- Consistency is NOT requested in words. You make the locked asset
(character / product / environment) an INPUT EDGE. The graph is the memory.
- Generation prompts in English.
WORKFLOW: first propose the node STRUCTURE for the scene type, then a
graph-aware prompt that references only nodes that exist. On a bad result,
do TARGETED diagnosis (which node, which param), not a generic rewrite.
LIMITS: headless edit is blind to the visual output, so validation stays with me.
scope=all is costly, so work per page or selection.
I'm not sure if I'm missing something, but when I go back and browse previous generations in a Space, I can't seem to see the prompt that was used for that specific generation.
Instead, the prompt field always shows the most recent prompt I used, even when I'm looking at much older generations.
Is this the intended behavior, or is there a way to view the exact prompt associated with each generation? It makes it difficult to compare results, troubleshoot changes, or learn from earlier generations when I can't see what input produced them.
Am I overlooking a setting, or is prompt history simply not stored per generation?
Hi guys!
I’m a motion designer, and I’m proposing to integrate AI into my clients' projects. But they think this is so easy and cheap to produce a video with AI, u should know that producing a video with consistent characters and environments is very complex and expensive (seedance started to cost a lot of credits). So it is very difficult to make a profit that makes it worth it from this.
How do u manage this? Like clients that ask for a lot of regeneration or clients that can't understand the real cost of money and time for this.
There's someone who’s really earning money? I mean enough to consider it as a job.
PS.
Just to clarify: my original question wasn’t really about which models people use.
I’m more interested in whether anyone is actually making solid margins from AI projects — enough to call it a real, worthwhile job.
And especially: how do you quote clients for this? How do you explain the real costs behind an AI project, like testing, credits, failed generations, revisions, cleanup, and consistency work?
Ciao a tutti. Ho 22 anni e sono un libero professionista, a inizio 2026 ho iniziato questo piano d’accumulo su questi 3 etf con l’obbiettivo di aumentare gradualmente l’importo settimanale al crescere del reddito. Sono ignorante e sto facendo tutto da solo quindi si accettano suggerimenti!
Attualmente corrisponde a circa il 15-20% delle mie entrate lorde e ho attualmente da parte in liquidità remunerata altri 15 mila euro (+4000 già messi sul portfolio in questi mesi), forse è un po troppo (?) ma preferisco rischiare ora che a 40 anni.
Questo è idealmente il mio primo tassello per iniziare a costruirmi da qui ai prossimi 10/20 anni un buon patrimonio per godermi la vita in serenità.
È un buon portfolio? Sono pochi solo 3 etf? Spero di essere stato esaustivo con le informazioni
